[Bugfix][MiniCPM-o] Build the Stage-0 duplex runtime at load time (#8146)

Signed-off-by: chickeyton <ngton2014@gmail.com>
Co-authored-by: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

tests/model_executor/models/minicpmo_4_5/duplex/test_stage0_runtime_load.py

@@ -0,0 +1,61 @@
+# SPDX-License-Identifier: Apache-2.0
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM-Omni project
+"""The MiniCPM-o 4.5 Stage-0 duplex runtime is built with the model, not in the first session."""
+
+from __future__ import annotations
+
+from types import SimpleNamespace
+
+import pytest
+import torch
+
+from vllm_omni.model_executor.models.minicpmo_4_5 import minicpmo_4_5_omni
+from vllm_omni.model_executor.models.minicpmo_4_5.duplex import compat
+from vllm_omni.model_executor.models.minicpmo_4_5.minicpmo_4_5_omni import (
+    MiniCPMO45OmniForConditionalGeneration,
+)
+
+pytestmark = [pytest.mark.core_model, pytest.mark.cpu]
+
+
+class _StageModel(torch.nn.Module):
+    def make_empty_intermediate_tensors(self):
+        return None
+
+
+@pytest.mark.parametrize(
+    ("model_stage", "session_mode", "expected_builds"),
+    [
+        ("llm", "duplex", 1),
+        ("llm", "turn", 0),
+        ("tts", "duplex", 0),
+    ],
+)
+def test_init_builds_duplex_runtime_only_for_duplex_thinker(
+    monkeypatch: pytest.MonkeyPatch,
+    model_stage: str,
+    session_mode: str,
+    expected_builds: int,
+) -> None:
+    monkeypatch.setattr(minicpmo_4_5_omni, "init_vllm_registered_model", lambda **kwargs: _StageModel())
+    monkeypatch.setattr(compat, "patch_minicpmo_remote_config", lambda config: None)
+    build_devices: list[torch.device] = []
+    monkeypatch.setattr(
+        MiniCPMO45OmniForConditionalGeneration,
+        "_duplex_data_plane_helper",
+        lambda self: build_devices.append(torch.get_default_device()),
+    )
+    vllm_config = SimpleNamespace(
+        model_config=SimpleNamespace(
+            hf_config=SimpleNamespace(),
+            multimodal_config=None,
+            model_stage=model_stage,
+            session_mode=session_mode,
+        )
+    )
+
+    # The loader constructs the model under the target-device context.
+    with torch.device("meta"):
+        MiniCPMO45OmniForConditionalGeneration(vllm_config=vllm_config)
+
+    assert build_devices == [torch.device("cpu")] * expected_builds

vllm_omni/model_executor/models/minicpmo_4_5/minicpmo_4_5_omni.py

@@ -144,6 +144,16 @@ class MiniCPMO45OmniForConditionalGeneration(nn.Module, SupportsMultiModal, Supp
         # embeddings and initializes request-local codec generation state.
         self.has_preprocess = self.model_stage in {"llm", "tts"}
 
+        if self.model_stage == "llm" and getattr(vllm_config.model_config, "session_mode", "turn") == "duplex":
+            # Build the Stage-0 duplex runtime (remote-code processor and
+            # tokenizer) with the model. Built lazily, it costs several seconds
+            # inside the first session's first audio unit, and the session then
+            # runs that far behind the real-time input stream. The loader
+            # constructs the model under the target-device context; the
+            # processor is CPU preprocessing, so keep its tensors on the CPU.
+            with torch.device("cpu"):
+                self._duplex_data_plane_helper()
+
     @cached_property
     def sampler(self):
         if hasattr(self.model, "sampler"):